query-resilience-serves-self-correcting-knowledge

OUT derived (depth 5)

All query access paths — interactive LLM synthesis, batch search, and compact summarization — degrade gracefully with deterministic output while operating against a knowledge base that actively self-corrects through contradiction resolution and staleness detection, ensuring degraded queries still return data from a consistency-maintained belief network

Summary

Even when the system is running in a reduced capacity — falling back to simpler search instead of full LLM-powered answers — the results still come from a knowledge base that is actively cleaning up its own contradictions and flagging stale information. This means degraded performance never silently serves bad data; the self-correction machinery runs independently of which query path is being used.

Justifications

SL — Deterministic query degradation (depth-4) ensures reliable access while self-correction (depth-4) ensures the accessed data is actively maintained — users get reliable reads against a self-healing store

Antecedents (all must be IN):

  • query-degradation-is-deterministic-across-all-access-paths — All information access paths degrade gracefully while maintaining deterministic output: interactive queries cascade through tiered modes (LLM synthesis → bounded tool loop → raw FTS5 search), structured reads self-heal missing indexes via derived FTS5 reconstruction, and all fallback paths produce deterministic sorted output.
  • complete-system-is-self-correcting — The system actively maintains its own consistency along two independent dimensions: the TMS core handles exceptional conditions (contradictions trigger deterministic resolution, propagation respects lifecycle state), while belief currency management detects and surfaces drift in source material — no inconsistency persists undetected or unresolved.

Dependents

These beliefs depend on this one:

Details